Behaviour
Our first module is the most
fundamental: behaviour. The aim of systems neuroscience is to figure out
the biological basis of thought and behaviour. This is done by
correlating biology -- notably, measuring or manipulating neural firing
rates -- with behaviour. That's how it is done even if we are more
interested internal reasoning, emotion, or motivation than in behaviour
itself: since we cannot yet directly measure thoughts, the human subject
or experimental animal's actions are always used as a proxy.
For a cognitive feature of interest, choosing the right behaviour to
train or measure is no simple task. To yield insightful experiments, the
behaviour must be the right combination of robust, naturalistic, and
truly reflective of the relevant thought process.
In this module, we wanted to find out to what degree rats are capable of
recognizing and manipulating objects. We know that rats excel in a
foraging task, in which they learn to navigate to a moving spot of light
in order to earn a sugary reward. This indicates that they are capable
of locating and navigating to sensory 'objects'. This, however, is a
very limited component of the kinds of interactions humans have with
objects, which culminates in the use and invention of complex tools.
Taking one step toward this extreme, we would like to know whether rats
can learn not just to navigate to an object but also to move it to a
target location. Specifically, we attempted to train our rats to push a
marble into a hole. Note that this is more complex than, say, pushing a
lever to earn a reward; pushing a lever simply requires a particular
motor act when the rat is in front of the lever. Getting a marble into a
hole, however, requires a distinct set of pushes depending on the
location of the marble and hole. This is a more flexible type of
behaviour; it requires an abstract understanding that the marble should
fall in the hole and that the rat is capable of causing that to happen.
Can rats easily learn this kind of object-oriented thinking? Can we
design a behavioural assay to effectively test that question? If yes to
both, can we look in the brain and discover the neurobiology behind this
simple model of object recognition and manipulation?
Here we attempt to create a behavioural assay to test how well rats can
recognize and manipulate objects. With this tricky task, nuanced
question, two rats and one week of training, we were doomed to
inconclusivity. Nonetheless, here are our methods, data, and
conclusions:
Materials & Methods
We trained our rat in 13 sessions of 30-50
minutes over 9 days, in a 46 x 80 cm behaviour rig.
The floor consists of two stacked rectangular acrylic boards: one with a
3-cm hole in the center to fit the 2.6-cm marble, and the other with
larger central hole of 20-cm diameter. Stacking these allowed us to
place three different 3D-printed circular sections in the middle
(schematic below). One was otherwise flat (most difficult); one was
sloped so that the marble would slide into the hole if it came within 10
cm of the center (medium-difficulty); and one was sloped across the
entire 20-cm diameter (least difficult). All pieces were designed in
Illustrator. See the schematic below for a representation of these
centerpieces. Due to the low friction of the acrylic board, we sanded
the board and marbles, and installed grooves along diameter-sections of
the most-sloped centerpiece. This made it easier for the marble to roll
into the hole instead of past the hole.
Left: Schematic of the flat, most
difficult (top), partially ramped, medium-difficulty (middle), and
fully ramped, least difficult (bottom) centerpieces, used in training.
Right: 3D rendering of the medium-difficulty centerpiece
The reward port, courtesy of Lory, contained an infrared beam and
detector and a strawberry Ribena juice dispenser. Whenever the rat
sticks its head into the reward area, the IR beam breaks, sending a
signal via an Arduino Uno to a computer. If the head poke occurs during
the appropriate part of the trial, a drop of juice is dispensed. The
marble detector similarly contained an IR beam that, when broken by a
marble falling through the hole, would send a signal to the computer.
The overhead camera, responsible for the movies embedded below, was an
IR camera positioned above the behaviour rig that recorded each training
session.
We coded all training automation and time-stamp-saving in Bonsai. Video
analysis was performed in MATLAB.
Behavioural training progressed through several stages. First, the rat
habituated to the beahviour rig and human handling over 2 days. Next,
over 6 days the rat was trained on an association between a beeping
sound (on for 50 seconds or until the rat arrived at the reward port,
and then off for 10-30 seconds; as training progressed the beeping was
reduced to 20 seconds on) and a juice reward in the reward port. During
this training, when the rat pokes its head in the reward port, it will
receive juice, but only if the beeping sound is on at that moment. This
was meant to allow us to later reward an action distant from the reward
port (pushing the marble) by turning on the beeping and allowing the rat
to collect reward soon after it pushes the marble. After entraining this
beep-juice association, we put the marbles in the arena.
The arena was now set up with 5-7 marbles placed around the outer
circumference of the centerpiece at the beginning of each session. The
plan at this point was as follows: reward any interaction with marbles
by turning on the beeping and reward availability; as soon as the rat
comes to regularly interact with marbles, decrease this manual rewarding
and only provide reward (automatically) when a marble falls through the
hole (of the least difficult centerpiece); as soon as the rat comes to
regularly push marbles into the least difficult hole, switch to the
medium-difficulty centerpiece; and as soon as the rat comes to regularly
push marbles into the medium-difficulty hole, switch to the most
difficult centerpiece. Finally, once the rat comes to regularly push
marbles into the most difficult hole, modify the parameters of the task
to test whether the rat has truly learned the abstract, object-oriented
thinking we attempted to entrain or if it instead learned a simpler rule
that also allowed it to effectively push marbles in the hole. For
example, we could place marbles in another arena adjoined to the one
with the hole by a narrow passage. It is relatively unlikely for the rat
to push marbles across the narrow passage unless it truly possessed the
goal of getting marbles into the hole.
Given just one rat and 3 days of training with marbles, this training
protocol did not progress to its culmination or yield a definitive
result. As shown below, we instead were able to make some observations
about our rat's particular habits during training (given an n=1, we
cannot generalize to other rats). We modified our reward protocol
accordingly, now providing chocolate treats by hand whenever the rat
interacted with the marble or pushed it into the hole. This more
effectively encouraged the rat to interact with the marble. The
experiment terminated at this stage of training.
Results
Days 1-6 (sessions 1 - 8) -- Beep-Juice-Association Training:
Over these sessions, the rat was trained on the above-described
beep-juice-association training protocol. Although avoiding falling
asleep, the rat only sporadically visited the reward port. After some
manual intervention -- dropping a trail of juice leading to the reward
port -- its interest in the port maginally increased. As this learning
occurred, we increased the inter-beep-interval (up to 30 sec) and
decreased the beeping duration (down to 20 sec). However, a robust
beeping-followed-by-running-to-collect-juice association never fully
formed.
Video and heat map showing the rat's
position over beep-juice-association training session #5. While
showing some interest in the reward port (bottom-middle region), the
rat spent more time in a corner (top-left region) and some time in the
middle (middle region).
Days 7 - 9 (sessions 9 - 13) -- Marble-Pushing Training:
Sessions 9-11 followed our original training protocol from Materials and
Methods. However, during these sessions the rat demonstrated a
surprising behavioural pattern that led us to modify our reward protocol
in sessions 11-12. At the beginning session 9 and 10, and sporadically
throughout the session, the rat demonstrated a high intrinsic interest
in the marbles. It interacted with them, mostly ignoring the beeping and
potential juice reward that followed. Indeed, with the least-difficult
centerpiece in place, the rat quickly pushed several marbles into the
hole, and then ignored the resulting reward availability at the port
(video below). This demonstrated that the rat had an intrinsic interest
in interacting with the marbles, that could potentially be shaped toward
specifically pushing them into the hole. It also showed that our
beep-juice association was inadequate for entraining this task.
Video: The rat had an untrained interest
in the marbles but lacked interest in the resulting juice
availability.
In sessions 11-12, we modified our reward
protocol accordingly, now providing chocolate treats by (Lory's) hand
whenever the rat interacted with the marble or pushed it into the hole.
As this required no effort on the rat's part beyond gulping up a
chocolate lowered to its mouth, it now collected its reward every time
the reward was earned. During session 11, the medium-difficulty
centerpiece was in place, and the rat was rewarded each time it pushed a
marble even slightly -- however, as the session progressed only pushed
toward the center of the arena were rewarded. During this session, the
rat pushed 6/7 marbles into the hole. In addition, when it pushed
marbles away from the center, it started to demonstrate a new behaviour:
stopping and looking upward, perhaps in anticipation of Lory's
sugar-bestowing hand (video below). During session 12, the most
difficult centerpiece was in place, and the rat was only rewarded for
pushing a marble toward the center. It pushed 4/5 marbles into the hole,
and demonstrated a sustained interest in interacting with the marbles.
Video 1: After having learned to interact
with marbles in order to receive a treat, we started to only reward
pushes toward the center. At this point, the rat paused and looked
upward after touching a marble or pushing it away from the hole,
presumably in expectation of chocolate; Video 2: An example of the rat
pushing the marble toward the center in order to earn a chocolate
treat -- the culmination of our training.
Conclusions
We were interested in finding out to what degree rats can learn
abstract, object-oriented tasks. The task we selected was to push
marbles placed around an arena into a small hole at the center of the
arena. Previous work has shown that rats can learn conceptually similar
and physically even more difficult tasks, such as basketball (video
below). Our goal was to optimize a training protocol for a simpler task;
once learned, our marble-pushing task could then be systematically
modified and extended in order to probe exactly what types of abstract
object-manipulation tasks rats are capable of. We did not get that far
after 8 days of training. However, we did make one key update to our
original materials and methods -- replacing the 5 days of ineffective
juice-beep-association training with a chocolate reward as soon as the
rat pushes the marble. With this change, our rat began to learn to push
marbles. We believe that some proportion of rats trained on this
protocol would learn to push marbles into a hole, and could then present
a rodent model system for studying the cognition of object manipulation.
Rats are capable of basic
object-manipulation tasks -- can we harness that as a model system in
the laboratory?
